Subsurface Defect Detection and Classification in 3D THz Images of Glass Fiber Reinforced Thermoplastic Based on 3D Convolutional Neural Network

This research proposes a 3D convolutional neural network (3D CNN) to classify subsurface defects in a glass fiber reinforced thermoplastic (GFRT) composite material inspected by a 3D THz imaging system. The 3D THz inspection system generates 3D volumetric data based on frequency modulated continuous wave (FMCW) technique operating in a frequency range of 499 GHz to 733 GHz. A set of 89 non-defective and defective GFRT samples is measured and data augmentation is performed to create a database for 3D CNN training. The results demonstrate that the 3D CNN achieves a mean test accuracy of 98.9%.